You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
feat: entropy-based secret detection for exception code variables
Add a last-resort entropy-based detector that redacts high-entropy,
secret-looking values (API keys, tokens, strong passwords) sitting in
innocuously-named code variables, after the existing name-pattern and
URL-credential checks.
- Known vendor key formats (OpenAI, Anthropic, AWS, Stripe, GitHub,
GitLab, Slack, Google, JWT, PEM private keys) are matched directly.
- Structured identifiers (UUIDs, Mongo ObjectIds, hashes), object reprs,
file paths and URLs are never flagged.
- Exposed as the `code_variables_detect_secrets` option (default True)
with a per-context override, threaded through client/contexts.
- Tighten the masking size caps to keep capture cost bounded.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Add entropy-based secret detection as a last-resort redaction step when capturing exception code variables. High-entropy, secret-looking values (API keys, tokens, strong passwords) sitting in innocuously-named variables are now redacted after the existing name-pattern and URL-credential checks. Well-known vendor key formats (OpenAI, Anthropic, AWS, Stripe, GitHub, GitLab, Slack, Google, JWT, PEM private keys) are matched directly, while structured identifiers (UUIDs, Mongo ObjectIds, hashes), object reprs, file paths and URLs are left intact. Adds the `code_variables_detect_secrets` option (default `True`) with a per-context override. Also tightens the code-variable size caps so capture stays bounded.
0 commit comments